ggplot2: further steps

(Partially abridged from r.statistics.co)

library(tidyverse)

Top ggplot2 Visualizations

An effective chart is one that:

1. Conveys the right information without distorting facts.
2. Is simple but elegant. It should not force you to think much in order to get it.
3. Aesthetics supports information rather that overshadow it.
4. Is not overloaded with information.

The list below sorts the visualizations based on its primary purpose. Primarily, there are 8 types of objectives you may construct plots. So, before you actually make the plot, try and figure what findings and relationships you would like to convey or examine through the visualization. Chances are it will fall under one (or sometimes more) of these 8 categories: Correlation, Deviation, Ranking, Distribution, Composition, Change, Groups and Spatial.

In this tutorial we’ll cover categories from Correlation to Composition, leaving Change, Groups and Spatial to the next lesson.

Correlation

The following plots help to examine how well correlated two variables are.

Scatterplot

The most frequently used plot for data analysis is undoubtedly the scatterplot. Whenever you want to understand the nature of relationship between two variables, invariably the first choice is the scatterplot.

It can be drawn using geom_point(). Additionally, geom_smooth which draws a smoothing line (based on loess) by default, can be tweaked to draw the line of best fit by setting method='lm'.

theme_set(theme_bw())  # global preset, bw theme
data("midwest", package = "ggplot2")
# midwest <- read.csv('http://goo.gl/G1K41K') # bkup data
# source

# Scatterplot
gg <- ggplot(midwest, aes(x = area, y = poptotal)) + geom_point(aes(col = state,
    size = popdensity)) + geom_smooth(method = "loess", se = F) +
    xlim(c(0, 0.1)) + ylim(c(0, 5e+05)) + labs(subtitle = "Area Vs Population",
    y = "Population", x = "Area", title = "Scatterplot", caption = "Source: midwest")

plot(gg)

Scatterplot With Encircling

When presenting the results, sometimes I would encirlce certain special group of points or region in the chart so as to draw the attention to those peculiar cases. This can be conveniently done using the geom_encircle() in the ggalt package.

Within geom_encircle(), set the data to a new dataframe that contains only the points (rows) or interest. Moreover, you can expand the curve so as to pass just outside the points. The color and size (thickness) of the curve can be modified as well.

library(ggalt)
midwest_select <- midwest %>% dplyr::filter(poptotal > 350000,
                                            poptotal <= 500000,
                                            area > 0.01,
                                            area < 0.1)

# Plot
ggplot(midwest, aes(x=area, y=poptotal)) + 
    geom_point(aes(col=state, size=popdensity)) + # draw points
    geom_smooth(method="loess", se=FALSE) + # draw smoothing line
    xlim(c(0, 0.1)) + 
    ylim(c(0, 500000)) + 
    geom_encircle(aes(x=area, y=poptotal), 
                  data=midwest_select, # filtered dataframe
                  color="red", 
                  size=2, 
                  expand=0.08) + # expand the curve a little bit outside the points
    labs(subtitle="Area Vs Population", 
         y="Population", 
         x="Area", 
         title="Scatterplot + Encircle", 
         caption="Source: midwest")

Jitter Plot

Let’s consider a new dataset now: I will use the mpg dataset to plot city mileage (cty) vs highway mileage (hwy).

data(mpg, package = "ggplot2")  # alternate source: 'http://goo.gl/uEeRGu')
theme_set(theme_bw())

g <- ggplot(mpg, aes(cty, hwy))

# Scatterplot
g + geom_point(size = 1) + geom_smooth(method = "lm", se = FALSE) +
    labs(subtitle = "mpg: city vs highway mileage", y = "hwy",
        x = "cty", title = "Scatterplot with overlapping points")

What we have here is a scatterplot of city and highway mileage in the mpg dataset. This scatterplot looks neat and gives a clear idea of how the city mileage (cty) and highway mileage (hwy) are well correlated.

But, this innocent looking plot is hiding something. Can you find out?

dim(mpg)
#> [1] 234  11

The original data has 234 data points but the chart seems to display fewer points! What has happened? This is because there are many overlapping points appearing as a single dot. The fact that both cty and hwy are integers in the source dataset made it all the more convenient to hide this detail. So just be extra careful the next time you make scatterplot with integers.

So how to handle this? There are few options. We can make a jitter plot with jitter_geom(). As the name suggests, the overlapping points are randomly jittered around its original position based on a threshold controlled by the width argument.

g + geom_jitter(width = 0.5, size = 1) + geom_smooth(method = "lm",
    se = FALSE) + labs(subtitle = "mpg: city vs highway mileage",
    y = "hwy", x = "cty", title = "Jittered Points")

More points are revealed now. The larger the jitter width, the more the points are moved (jittered) from their original position.

Counts Chart

The second option to overcome the problem of data points overlap is to use what is called a counts chart. Wherever there is more points overlap, the size of the circle gets bigger.

g + geom_count(col = "tomato3", show.legend = FALSE) + labs(subtitle = "mpg: city vs highway mileage",
    y = "hwy", x = "cty", title = "Counts Plot")

By default, geom_count() automatically inserts a legend for the circle sizes:

g + geom_count(col = "tomato3") + labs(subtitle = "mpg: city vs highway mileage",
    y = "hwy", x = "cty", title = "Counts Plot")

Bubble plot

While a scatterplot lets you compare the relationship between two continuous variables, a bubble chart serves well if you want to understand the relationship within the underlying groups, based on:

1. A categorical variable (by changing the color), and
2. Another continuous variable (by changing the size of points).

In simpler words, bubble charts are more suitable if you have 4-Dimensional data where two of them are numeric (X and Y) and one other categorical (color) and another numeric variable (size).

In the following example, we display the city mileage (cty) versus engine displacement (displ), encoding information about manufacturer as color and about highway consumption (hwy) as size.

The bubble chart clearly distinguishes the range of displ between the manufacturers and how the slope of best-fit lines varies, providing a better visual comparison between the groups.

mpg_select <- mpg %>%
    dplyr::filter(manufacturer %in% c("audi", "ford", "honda",
        "hyundai"))

g <- ggplot(mpg_select, aes(displ, cty)) + labs(subtitle = "mpg: City Mileage vs. Displacement",
    title = "Bubble chart")

g + geom_jitter(aes(col = manufacturer, size = hwy)) + geom_smooth(aes(col = manufacturer),
    method = "lm", se = F)

Marginal Histogram / Boxplot

If you want to show the relationship as well as the distribution in the same chart, use the marginal histogram. It has a histogram of the X and Y variables at the margins of the scatterplot.

This can be implemented using the ggMarginal() function from the ggExtra package. Apart from a histogram, you could choose to draw a marginal boxplot or density plot by setting the respective type option.

library(ggExtra)

g <- ggplot(mpg, aes(cty, hwy)) + geom_count(show.legend = FALSE) +
    geom_smooth(method = "lm", se = F)

ggMarginal(g, type = "histogram", fill = "transparent")

ggMarginal(g, type = "boxplot", fill = "transparent")

ggMarginal(g, type = "density", fill = "transparent")

ggMarginal(g, type = "densigram")  # density + histogram

Correlogram

Correlograms let you examine the correlation of multiple continuous variables present in the same dataframe. This is conveniently implemented using the ggcorrplot package.

We explore an example on the mtcars dataset, containing fuel consumption and 10 aspects of car design and performance for 32 cars (models from 1973-1974).

library(ggcorrplot)

data(mtcars)
dim(mtcars)
#> [1] 32 11
# compute the correlation matrix
corr <- round(cor(mtcars), 1)

# plot
ggcorrplot(corr, 
           hc.order = FALSE, # order the corr. matrix by hierarchical clustering
           type = "lower", 
           lab = TRUE, # add corr. coefficients
           lab_size = 3, 
           method="circle", 
           colors = c("tomato2", "white", "springgreen3"), # colors for low, mid, high correlation values
           title="Correlogram of mtcars", 
           ggtheme=theme_bw)

Deviation

Compare variation in values between small number of items (or categories) with respect to a fixed reference.

Diverging bars

“Diverging bars” is a kind of bar chart that can handle both negative and positive values. This can be implemented by a smart tweak with geom_bar(). But the usage of geom_bar() can be quite confusing, because it can be used to make a bar chart as well as a histogram.

By default, geom_bar() has the stat argument set to count. That means, when you provide just a continuous X variable (and no Y variable), it tries to make a histogram out of the data. We saw an example of this in the Lab3 slides.

In order to make a bar chart create bars instead of a histogram, you need to do two things:

1. Set stat=identity (that means, plot the values as they are)
2. Provide both x and y inside aes(), where x is either character or factor and y is numeric.

In order to make sure you get diverging bars instead of just bars, make sure your categorical variable has 2 categories that change values at a certain threshold of the continuous variable. In the below example, the mpg from mtcars dataset is normalized by computing the z score. Those vehicles with \(\textit{mpg}\geq 0\) are marked green and those below are marked red.

data("mtcars")
# data prep
mtcars <- tibble::rownames_to_column(mtcars, var="car name") %>% # create new column for car names
    mutate(mpg_z=round(scale(mpg), 2), # compute normalized mpg
           mpg_type=ifelse(mpg_z < 0, "below", "above"), # above / below avg flag
    ) %>% 
    arrange(mpg_z)

mtcars$`car name` <- factor(mtcars$`car name`, levels = mtcars$`car name`)  # convert to factor to retain sorted order in plot.

# diverging bars
ggplot(mtcars, aes(x=`car name`, y=mpg_z, label=mpg_z)) + 
    geom_bar(stat="identity", aes(fill=mpg_type), width=.5)  +
    scale_fill_manual(name="Mileage", 
                      labels = c("Above Average", "Below Average"), 
                      values = c("above"="#00ba38", "below"="#f8766d")) + 
    labs(subtitle="Normalized mileage from mtcars", 
         title= "Diverging Bars") + 
    coord_flip() +
    theme_bw()

Diverging Lollipop Chart

Lollipop chart conveys the same information as bar charts and diverging bars, except that it looks more modern. Instead of geom_bar, I use geom_point and geom_segment to get the lollipops right. Now let’s draw a lollipop using the same data I prepared in the previous example of diverging bars.

geom_segment draws a straight line between points (x, y) and (xend, yend).

ggplot(mtcars, aes(x = `car name`, y = mpg_z, label = mpg_z)) +
    geom_point(stat = "identity", fill = "black", size = 6) +
    geom_segment(aes(y = 0, x = `car name`, yend = mpg_z, xend = `car name`),
        color = "black") + geom_text(color = "white", size = 2) +
    labs(title = "Diverging Lollipop Chart", subtitle = "Normalized mileage from mtcars: Lollipop") +
    ylim(-2.5, 2.5) + coord_flip() + theme_bw()

Diverging Dot Plot

Dot plots convey similar information. The principles are same as what we saw in Diverging bars, except that only points are used. The below example uses the same data prepared in the diverging bars example.

ggplot(mtcars, aes(x = `car name`, y = mpg_z, label = mpg_z)) +
    geom_point(stat = "identity", aes(col = mpg_type), size = 6) +
    scale_color_manual(name = "Mileage", labels = c("Above Average",
        "Below Average"), values = c(above = "#00ba38", below = "#f8766d")) +
    geom_text(color = "white", size = 2) + labs(title = "Diverging Dot Plot",
    subtitle = "Normalized mileage from 'mtcars': Dotplot") +
    ylim(-2.5, 2.5) + coord_flip() + theme_bw()

Area Chart

Area charts are typically used to visualize how a particular metric (such as % returns from a stock) performed compared to a certain baseline. Other types of %returns or %change data are also commonly used. The geom_area() implements this.

data("economics", package = "ggplot2")
glimpse(economics)
#> Rows: 574
#> Columns: 6
#> $ date     <date> 1967-07-01, 1967-08-01, 1967-09-01, 1967-10-01, 1967-11-01, …
#> $ pce      <dbl> 506.7, 509.8, 515.6, 512.2, 517.4, 525.1, 530.9, 533.6, 544.3…
#> $ pop      <dbl> 198712, 198911, 199113, 199311, 199498, 199657, 199808, 19992…
#> $ psavert  <dbl> 12.6, 12.6, 11.9, 12.9, 12.8, 11.8, 11.7, 12.3, 11.7, 12.3, 1…
#> $ uempmed  <dbl> 4.5, 4.7, 4.6, 4.9, 4.7, 4.8, 5.1, 4.5, 4.1, 4.6, 4.4, 4.4, 4…
#> $ unemploy <dbl> 2944, 2945, 2958, 3143, 3066, 3018, 2878, 3001, 2877, 2709, 2…
economics

# Compute %Returns
economics$returns_perc <- c(0, diff(economics$psavert)/economics$psavert[-length(economics$psavert)])
head(economics$returns_perc)
#> [1]  0.000000000  0.000000000 -0.055555556  0.084033613 -0.007751938
#> [6] -0.078125000

# Create break points and labels for axis ticks
brks <- economics$date[seq(1, length(economics$date), 12)]
lbls <- lubridate::year(brks)

# plot the 1st 100 observations
ggplot(economics[1:100, ], aes(date, returns_perc)) + geom_area() +
    scale_x_date(breaks = brks, labels = lbls) + labs(title = "Area Chart",
    subtitle = "Percentage Returns for Personal Savings", y = "% Returns for Personal savings",
    caption = "Source: economics dataset") + theme_bw() + theme(axis.text.x = element_text(angle = 90))

Ranking

A ranking plot is used to compare the position or performance of multiple items with respect to each other. Actual values matter somewhat less than the ranking.

Ordered Bar Chart

This is a Bar Chart that is ordered by the Y axis variable. Just sorting the dataframe by the variable of interest is not enough to order the bar chart: in order for the bar chart to retain the order of the rows, the X axis variable (i.e., the categories) has to be converted into a factor.

Let’s plot the mean city mileage for each manufacturer from the mpg dataset. First, aggregate the data and sort it before you draw the plot. Finally, the X variable is converted to a factor.

# data prep: group mean city mileage by manufacturer.
cty_mpg <- mpg %>%
    group_by(make = manufacturer) %>%
    summarise(mileage = mean(cty))
cty_mpg <- arrange(cty_mpg, mileage)  # sort
cty_mpg$make <- factor(cty_mpg$make, levels = cty_mpg$make)  # refactor to retain the order in plot.
head(cty_mpg, 4)

# Draw plot
ggplot(cty_mpg, aes(x = make, y = mileage)) + geom_bar(stat = "identity",
    width = 0.5, fill = "tomato3") + labs(title = "Ordered Bar Chart",
    subtitle = "Make Vs Avg. Mileage", caption = "source: mpg") +
    theme_bw() + theme(axis.text.x = element_text(angle = 65,
    vjust = 0.6))

Lollipop Chart

Lollipop charts convey the same information as bar charts. By reducing the thick bars into thin lines, they reduce the clutter and lay more emphasis on the value. They look nice and modern.

ggplot(cty_mpg, aes(x = make, y = mileage)) + geom_point(size = 3) +
    geom_segment(aes(x = make, xend = make, y = 0, yend = mileage)) +
    labs(title = "Lollipop Chart", subtitle = "Make Vs Avg. Mileage",
        caption = "source: mpg") + theme_bw() + theme(axis.text.x = element_text(angle = 65,
    vjust = 0.6))

Dot Plot

Dot plots are very similar to lollipops, except they don’t have segments and they are flipped to horizontal position. This chart emphasizes more the rank ordering of items with respect to actual values and how far apart are the entities with respect to each other.

ggplot(cty_mpg, aes(x=make, y=mileage)) + 
    geom_point(col="tomato2", size=3) + # draw points
    geom_segment(aes(x=make, 
                     xend=make, 
                     y=min(mileage), 
                     yend=max(mileage)), 
                 linetype="dashed", # draw dashed lines
                 size=0.1) +   
    labs(title="Dot Plot", 
         subtitle="Make Vs Avg. Mileage", 
         caption="source: mpg") +  
    coord_flip() +
    theme_classic()

Slope Chart

Slope charts are an excellent way of comparing the positional placements between two points on time. At the moment, there is no built-in function to construct this. The following code is a starting point to guide you on how to approach this.

library(scales)

# data prep
dataf <- read.csv("https://raw.githubusercontent.com/selva86/datasets/master/gdppercap.csv")
colnames(dataf) <- c("continent", "1952", "1957")
# prepare labels
left_label <- paste(dataf$continent, round(dataf$`1952`), sep=", ")
right_label <- paste(dataf$continent, round(dataf$`1957`), sep=", ")
dataf <- dataf %>% mutate(class=ifelse(`1957` - `1952` < 0, "red", "green"))

p <- ggplot(dataf) + geom_segment(aes(x=1, xend=2, y=`1952`, yend=`1957`, col=class), size=.75, show.legend=F) + 
    geom_vline(xintercept=1, linetype="dashed", size=.1) + 
    geom_vline(xintercept=2, linetype="dashed", size=.1) +
    scale_color_manual(labels = c("Up", "Down"), 
                       values = c("green"="#00ba38", "red"="#f8766d")) +  # color of lines
    labs(x="", y="Mean GdpPerCap") +  # Axis labels
    xlim(.5, 2.5) + ylim(0,(1.1*(max(dataf$`1952`, dataf$`1957`)))) +
    theme_classic()

# intermediate product
print(p)

# add texts
p <- p + geom_text(label=left_label, y=dataf$`1952`, x=rep(1, NROW(dataf)), hjust=1.1, size=3.5)
p <- p + geom_text(label=right_label, y=dataf$`1957`, x=rep(2, NROW(dataf)), hjust=-0.1, size=3.5)
p <- p + geom_text(label="Time 1", x=1, y=1.1*(max(dataf$`1952`, dataf$`1957`)), hjust=1.2, size=5)  # title
p <- p + geom_text(label="Time 2", x=2, y=1.1*(max(dataf$`1952`, dataf$`1957`)), hjust=-0.1, size=5)  # title

# Minify theme
p + theme(panel.background = element_blank(), 
          panel.grid = element_blank(),
          axis.ticks = element_blank(),
          axis.text.x = element_blank(),
          panel.border = element_blank(),
          plot.margin = unit(c(1,2,1,2), "cm"))

Dumbbell Plot

Dumbbell charts are a great tool if you wish to:

1. Visualize relative positions (like growth and decline) between two points in time.
2. Compare distances between two categories.

In order to get the correct ordering of the dumbbells, the Y variable should be a factor and the levels of the factor variable should be in the same order as it should appear in the plot - as we already did for the diverging bars and the ordered bar charts.

library(ggalt)

health <- read.csv("https://raw.githubusercontent.com/selva86/datasets/master/health.csv")
health$Area <- factor(health$Area, levels = as.character(health$Area))  # for the correct ordering of the dumbbells

ggplot(health, aes(x = pct_2014, xend = pct_2013, y = Area, group = Area)) +
    geom_dumbbell(color = "#a3c4dc", size = 0.75, colour_xend = "#0e668b") +
    scale_x_continuous(label = scales::percent) + labs(x = NULL,
    y = NULL, title = "Dumbbell Chart", subtitle = "Pct Change: 2013 vs 2014",
    caption = "Source: https://github.com/hrbrmstr/ggalt") +
    theme_classic() + theme(plot.title = element_text(hjust = 0.5,
    face = "bold"), plot.background = element_rect(fill = "#f7f7f7"),
    panel.background = element_rect(fill = "#f7f7f7"), panel.grid.minor = element_blank(),
    panel.grid.major.y = element_blank(), panel.grid.major.x = element_line(),
    axis.ticks = element_blank(), legend.position = "top", panel.border = element_blank())

Distribution

Use a distribution plot when you have lots and lots of data points and want to study where and how the data points are distributed.

Histogram

By default, if only one variable is supplied, geom_bar() tries to calculate the count. In order for it to behave like a bar chart, the stat=identity option has to be set and x and y values must be provided.

• Histogram on a continuous variable Histogram on a continuous variable can be accomplished using either geom_bar() or geom_histogram(). When using geom_histogram(), you can control the number of bars using the bins option. Else, you can set the range covered by each bin using binwidth. The value of binwidth is on the same scale as the continuous variable on which the histogram is built. Since geom_histogram allows you to control both the number of bins and binwidth, it is the preferred option to create a histogram on continuous variables.

theme_set(theme_classic()) # set the theme beforehand

# histogram on a continuous (numeric) variable
g <- ggplot(mpg, aes(displ)) + scale_fill_brewer(palette = "Spectral")

g + geom_histogram(aes(fill=class), 
                   binwidth = .1, # change binwidth
                   col="black", 
                   size=.1) +  
    labs(title="Histogram with Auto Binning", 
         subtitle="Engine Displacement across Vehicle Classes")  

g + geom_histogram(aes(fill=class), 
                   bins=5, # change number of bins
                   col="black", 
                   size=.1) +
  labs(title="Histogram with Fixed Bins", 
       subtitle="Engine Displacement across Vehicle Classes") 

• Histogram on a categorical variable Histogram on a categorical variable would result in a frequency chart showing bars for each category. By adjusting width, you can adjust the thickness of the bars.

theme_set(theme_classic())

# Histogram on a Categorical variable
g <- ggplot(mpg, aes(manufacturer))

g + geom_bar(aes(fill = class), width = 0.5) + theme(axis.text.x = element_text(angle = 65,
    vjust = 0.6)) + labs(title = "Histogram on Categorical Variable",
    subtitle = "Manufacturer across Vehicle Classes")

Density plot

theme_set(theme_classic())

g <- ggplot(mpg, aes(cty))

g + geom_density(aes(fill = factor(cyl)), alpha = 0.8) + labs(title = "Density plot",
    subtitle = "City Mileage Grouped by Number of cylinders",
    caption = "Source: mpg", x = "City Mileage", fill = "# Cylinders")

Box Plot

The Box plot (boxplot, box-and-whisker plot) is an excellent tool to explore distributions. It can also show the distributions within multiple groups, along with the median, range and outliers (if any).

The dark line inside the box represents the median. The top of box is the 3rd quartile and the bottom is the 1st quartile. The end points of the lines (a.k.a. “whiskers”) are at a distance of 1.5*IQR, where IQR (Inter Quartile Range) is the distance between 1st and 3rd quartiles (25th and 75th percentiles). The points outside the whiskers are marked as dots and are usually considered as extreme points (outliers).

In ggplot, you draw a boxplot adding the geometry geom_boxplot().

theme_set(theme_classic())

g <- ggplot(mpg, aes(class, cty))

g + geom_boxplot(fill = "plum") + labs(title = "Box plot", subtitle = "City Mileage grouped by Class of vehicle",
    caption = "Source: mpg", x = "Class of Vehicle", y = "City Mileage")

Setting varwidth=TRUE adjusts the width of the boxes to be proportional to the number of observation it contains.

g + geom_boxplot(varwidth = TRUE, fill = "plum") + labs(title = "Box plot",
    subtitle = "City Mileage grouped by Class of vehicle", caption = "Source: mpg",
    x = "Class of Vehicle", y = "City Mileage")

You can easily obtain a grouped box plot by stratifying on a factor variable:

g <- ggplot(mpg, aes(class, cty))
g + geom_boxplot(aes(fill = factor(cyl))) + theme(axis.text.x = element_text(angle = 65,
    vjust = 0.6)) + labs(title = "Box plot", subtitle = "City Mileage grouped by Class of vehicle",
    caption = "Source: mpg", x = "Class of Vehicle", y = "City Mileage")

Dot + Box Plot

On top of the information provided by a box plot, the dot plot can provide more clear information in the form of summary statistics by each group. The dots are staggered such that each dot represents one observation. So, in the below chart, the number of dots for a given manufacturer will match the number of rows of that manufacturer in source data.

theme_set(theme_bw())

g <- ggplot(mpg, aes(manufacturer, cty))

g + geom_boxplot() + geom_dotplot(binaxis = "y", stackdir = "center",
    dotsize = 0.5, fill = "red") + theme(axis.text.x = element_text(angle = 65,
    vjust = 0.6)) + labs(title = "Box plot + Dot plot", subtitle = "City Mileage vs Class: Each dot represents 1 row in source data",
    caption = "Source: mpg", x = "Class of Vehicle", y = "City Mileage")

A variant of this representation is obtained by jittering the dots, like in the following example:

g + geom_boxplot() + geom_point(position = position_jitter(width = 0.2),
    size = 1, color = "red") + theme(axis.text.x = element_text(angle = 65,
    vjust = 0.6)) + labs(title = "Box plot + Dot plot", subtitle = "City Mileage vs Class: Each dot represents 1 row in source data",
    caption = "Source: mpg", x = "Class of Vehicle", y = "City Mileage")

Wait… the outliers are plotted twice: by geom_boxplot() and by geom_point(). As a workaround, we switch them off in geom_boxplot() with outlier.color=NA:

g + geom_boxplot(outlier.color = NA) + geom_point(position = position_jitter(width = 0.2),
    size = 1, color = "red") + theme(axis.text.x = element_text(angle = 65,
    vjust = 0.6)) + labs(title = "Box plot + Dot plot", subtitle = "City Mileage vs Class: Each dot represents 1 row in source data",
    caption = "Source: mpg", x = "Class of Vehicle", y = "City Mileage")

Tufte Boxplot

Tufte box plot, provided by the ggthemes package, is inspired by the works of Edward Tufte: it is just a box plot made minimal and visually appealing.

library(ggthemes)
theme_set(theme_tufte())

g <- ggplot(mpg, aes(manufacturer, cty))

g + geom_tufteboxplot() + theme(axis.text.x = element_text(angle = 65,
    vjust = 0.6)) + labs(title = "Tufte Styled Boxplot", subtitle = "City Mileage grouped by Class of vehicle",
    caption = "Source: mpg", x = "Class of Vehicle", y = "City Mileage")

Violin Plot

A violin plot is similar to a box plot but it shows the density within groups. It does not provide as much info as a box plot. You can draw it using geom_violin().

theme_set(theme_bw())

g <- ggplot(mpg, aes(class, cty))

g + geom_violin() + labs(title = "Violin plot", subtitle = "City Mileage vs Class of vehicle",
    caption = "Source: mpg", x = "Class of Vehicle", y = "City Mileage")

Compare it with a box plot of the same data:

theme_set(theme_bw())

g <- ggplot(mpg, aes(class, cty))

g + geom_boxplot() + labs(title = "Violin plot", subtitle = "City Mileage vs Class of vehicle",
    caption = "Source: mpg", x = "Class of Vehicle", y = "City Mileage")

Population Pyramid

Population pyramids offer a unique way of visualizing how much population or what percentage of population fall under a certain category. The below pyramid is an excellent example of how many users are retained at each stage of a email marketing campaign funnel.

options(scipen = 999)  # turns of scientific notations like 1e+40

# get data
email_campaign_funnel <- read.csv("https://raw.githubusercontent.com/selva86/datasets/master/email_campaign_funnel.csv")

head(email_campaign_funnel)

# X axis breaks 
brks <- seq(-15000000, 15000000, 5000000)
# X axis labels
lbls <- paste0(as.character(c(seq(15, 0, -5), seq(5, 15, 5))), "m")

# pyramid
ggplot(email_campaign_funnel, aes(x = Stage, y = Users, fill = Gender)) + # Fill column
    geom_bar(stat = "identity", width = .6) +  # draw the bars
    scale_y_continuous(breaks = brks,   # Breaks
                       labels = lbls) + # Labels
    coord_flip() +  # Flip axes
    labs(title="Email Campaign Funnel") +
    theme_tufte() +  # Tufte theme from ggthemes
    theme(plot.title = element_text(hjust = .5), # Center plot title
          axis.ticks = element_blank()) +
    scale_fill_brewer(palette = "Dark2")  # Color palette

Composition

Waffle Chart

Waffle charts is a nice way of showing the categorical composition of the total population. Though there is no direct function, it can be articulated by smartly maneuvering ggplot2 using geom_tile(). The below template should help you create your own waffle.

var <- mpg$class  # categorical data 
table(var)  # original category distribution
#> var
#>    2seater    compact    midsize    minivan     pickup subcompact        suv 
#>          5         47         41         11         33         35         62
# data prep
nrows <- 10  # our waffle chart will be a 10x10 square
dataf <- expand.grid(y = 1:nrows, x = 1:nrows)
categ_table <- round(table(var) * ((nrows * nrows)/(length(var))))  # transform the category distribution so that the counts sum up to 100
categ_table
#> var
#>    2seater    compact    midsize    minivan     pickup subcompact        suv 
#>          2         20         18          5         14         15         26
# > 2seater compact midsize minivan pickup subcompact suv >
# 2 20 18 5 14 15 26
sum(categ_table)
#> [1] 100

dataf$category <- factor(rep(names(categ_table), categ_table))
# NOTE: if sum(categ_table) is not 100 (i.e. nrows^2), it
# will need adjustment to make the sum to 100.

# waffle chart
ggplot(dataf, aes(x = x, y = y, fill = category)) + geom_tile(color = "black",
    size = 0.5) + scale_x_continuous(expand = c(0, 0)) + scale_y_continuous(expand = c(0,
    0), trans = "reverse") + scale_fill_brewer(palette = "Set3") +
    labs(title = "Waffle Chart", subtitle = "'Class' of vehicles",
        caption = "Source: mpg") + theme(panel.border = element_rect(size = 2),
    plot.title = element_text(size = rel(1.2)), axis.text = element_blank(),
    axis.title = element_blank(), axis.ticks = element_blank(),
    legend.title = element_blank(), legend.position = "right")

Pie Chart

Pie chart, a classic way of showing compositions, is equivalent to the waffle chart in terms of the information conveyed. But it is slightly tricky to implement in ggplot2 using the coord_polar(). First, we create a standard bar chart and then we change to polar coordinates to make it a pie chart:

theme_set(theme_classic())

# Source: Frequency table
dataf <- as.data.frame(table(mpg$class))
colnames(dataf) <- c("class", "freq")

pie <- ggplot(dataf, aes(x = "", y = freq, fill = factor(class))) +
    geom_bar(width = 1, stat = "identity") + theme(axis.line = element_blank(),
    plot.title = element_text(hjust = 0.5)) + labs(fill = "class",
    x = NULL, y = NULL, title = "Pie Chart of class", caption = "Source: mpg")

# what we got so far
print(pie)

# transform to polar coordinates
pie + coord_polar(theta = "y", start = 0)

We are almost there! Now we would like to get rid of the original axis ticks and labels: we can do that with some polishing with the theme() function afterwards:

pie + coord_polar(theta = "y", start = 0) + theme(axis.ticks = element_blank(),
    axis.text = element_blank(), axis.title = element_blank(),
    panel.grid = element_blank())

Treemap

In a treemap, each tile represents a single observation, with the area of the tile proportional to a variable. Let’s start by drawing a treemap with each tile representing a G-20 country. The area of the tile will be mapped to the country’s GDP, and the tile’s fill colour mapped to its HDI (Human Development Index).

The treemapify package provides the basic geom for this purpose, geom_treemap().

library(treemapify)
ggplot(G20, aes(area = gdp_mil_usd, fill = hdi)) + geom_treemap()

This plot isn’t very useful without the knowing what country is represented by each tile.

geom_treemap_text can be used to add a text label to each tile. It uses the ggfittext package to resize the text so it fits the tile. In addition to standard text formatting aesthetics you would use in geom_text, like fontface or colour, we can pass additional options specific for ggfittext: for example, we can place the text in the middle of the tile with place="centre".

ggplot(G20, aes(area = gdp_mil_usd, fill = hdi, label = country)) +
    geom_treemap() + geom_treemap_text(fontface = "italic", colour = "white",
    place = "centre")

We can expand the tile text to fill as much of the tile as possible with grow=TRUE:

ggplot(G20, aes(area = gdp_mil_usd, fill = hdi, label = country)) +
    geom_treemap() + geom_treemap_text(fontface = "italic", colour = "white",
    place = "centre", grow = TRUE)

Note that some tiles in the top right corner may appear to have no labels (unless you enlarge the plot window). geom_treemap_text will hide text labels that cannot fit a tile without being shrunk below a minimum size, by default 4 points. This can be adjusted with the min.size argument.

geom_treemap supports subgrouping of tiles within a treemap by passing a subgroup aesthetic. Let’s subgroup the countries by region, draw a border around each subgroup with geom_treemap_subgroup_border, and label each subgroup with geom_treemap_subgroup_text.

geom_treemap_subgroup_text takes the same arguments for text placement and resizing as geom_treemap_text.

ggplot(G20, aes(area = gdp_mil_usd, fill = hdi, label = country,
    subgroup = region)) + geom_treemap() + geom_treemap_subgroup_border() +
    geom_treemap_subgroup_text(place = "centre", grow = T, alpha = 0.5,
        colour = "black", fontface = "italic", min.size = 0) +
    geom_treemap_text(colour = "white", place = "topleft", reflow = T)

Up to three nested levels of subgrouping are supported with the subgroup2 and subgroup3 aesthetics. Borders and text labels for these subgroups can be drawn with geom_treemap_subgroup2_border, etc.

Note that ggplot2 draws plot layers in the order that they are added. This means it is possible to accidentally hide one layer of subgroup borders with another. Usually, it’s best to add the border layers in order from deepest to shallowest, i.e. geom_treemap_subgroup3_border then geom_treemap_subgroup2_border then geom_treemap_subgroup_border.

ggplot(G20, aes(area = 1, label = country, subgroup = hemisphere,
    subgroup2 = region, subgroup3 = econ_classification)) + geom_treemap() +
    geom_treemap_subgroup3_border(colour = "blue", size = 1) +
    geom_treemap_subgroup2_border(colour = "white", size = 3) +
    geom_treemap_subgroup_border(colour = "red", size = 5) +
    geom_treemap_subgroup_text(place = "middle", colour = "red",
        alpha = 0.5, grow = T) + geom_treemap_subgroup2_text(colour = "white",
    alpha = 0.5, fontface = "italic") + geom_treemap_subgroup3_text(place = "top",
    colour = "blue", alpha = 0.5) + geom_treemap_text(colour = "white",
    place = "middle", reflow = T)

Bar Chart

A bar chart (geom_bar()) can be drawn from a categorical column variable or from a separate frequency table. By adjusting width, you can adjust the thickness of the bars. Remember that if your data source is a frequency table, that is, if you don’t want ggplot to compute the counts, you need to set the stat=identity inside the geom_bar().

# data prep: frequency table
freqtable <- table(mpg$manufacturer)
dataf <- as.data.frame.table(freqtable) %>%
    rename(manufacturer = Var1)
head(dataf)

theme_set(theme_classic())
g <- ggplot(dataf, aes(manufacturer, Freq))
g + geom_bar(stat = "identity", width = 0.5, fill = "tomato2") +
    labs(title = "Bar Chart", subtitle = "Manufacturer of vehicles",
        caption = "Source: Frequency of Manufacturers from 'mpg' dataset") +
    theme(axis.text.x = element_text(angle = 65, vjust = 0.6))

The frequency can be computed directly from a column variable as well. In this case, only X is provided and stat=identity is not set. While we are at it, we create a stacked bar chart showing the breakdown of car class.

# From on a categorical column variable directly
g <- ggplot(mpg, aes(manufacturer))
g + geom_bar(aes(fill=class), width = 0.5) + # fill by class
    theme(axis.text.x = element_text(angle=65, vjust=0.6)) +
    labs(title="Categorywise Bar Chart", 
         subtitle="Manufacturer of vehicles", 
         caption="Source: Manufacturers from 'mpg' dataset")